# DeepSeek releases V4 Pro and V4 Flash preview models at fraction of frontier prices

_Friday, April 24, 2026 at 4:23 AM EDT · AI · Latest · Tier 1 — Major_

![DeepSeek releases V4 Pro and V4 Flash preview models at fraction of frontier prices — Primary](https://static.simonwillison.net/static/2026/deepseek-v4-card.jpg)

Chinese AI lab DeepSeek has released two preview models, DeepSeek-V4-Pro and DeepSeek-V4-Flash, both featuring a 1 million token context window and a Mixture of Experts architecture. DeepSeek published the models under the standard MIT license.

DeepSeek-V4-Pro is 1.6 trillion total parameters with 49 billion active, making it the largest open weights model available. It is larger than Kimi K2.6 at 1.1 trillion parameters and GLM-5.1 at 754 billion parameters. The Pro model is 865GB on Hugging Face. DeepSeek-V4-Flash is 284 billion total parameters with 13 billion active and is 160GB.

The company is charging $1.74 per million input tokens and $3.48 per million output tokens for Pro, and $0.14 per million input tokens and $0.28 per million output tokens for Flash. DeepSeek-V4-Flash is the cheapest of the small models, beating OpenAI's GPT-5.4 Nano. DeepSeek-V4-Pro is the cheapest of the larger frontier models.

DeepSeek said in its research paper that the models achieve significant efficiency gains. In a 1 million token context scenario, DeepSeek-V4-Pro attains only 27 percent of the single-token FLOPs and 10 percent of the KV cache size relative to DeepSeek-V3.2. DeepSeek-V4-Flash achieves only 10 percent of the single-token FLOPs and 7 percent of the KV cache size compared with DeepSeek-V3.2.

DeepSeek's self-reported benchmarks show the Pro model competitive with frontier models from Gemini, OpenAI and Anthropic. The company noted that DeepSeek-V4-Pro-Max demonstrates superior performance relative to GPT-5.2 and Gemini-3.0-Pro on standard reasoning benchmarks, but falls marginally short of GPT-5.4 and Gemini-3.1-Pro, suggesting a developmental trajectory that trails state-of-the-art frontier models by approximately 3 to 6 months.

## Sources

- [Simon Willison's Weblog](https://simonwillison.net/2026/Apr/24/deepseek-v4/)
- [Techmeme](http://www.techmeme.com/260424/p10#a260424p10)

---
Canonical: https://techandbusiness.org/newswire/nboQQLUk2FOYJmEHhYETXG
Retrieved: 2026-04-24T13:02:54.312Z
Publisher: Tech & Business (techandbusiness.org)